A domain-independent knowledge-based framework is being developed, specific to the task of abstraction of higher-level, interval-based concepts from time-stamped data in a context- sensitive manner. The inference structure used is knowledge- based temporal-abstraction, which decomposes the temporal- abstraction task into five subtasks, each solved by a temporal- abstraction mechanism. The research provides a formal method for specification of temporal-abstraction knowledge and for facilitating its acquisition, maintenance, reuse, and sharing. A representation of the knowledge in the web-based Ontolingua language also is evaluated. The temporal-abstraction mechanisms are being extended and validated within the RESUME system. An automated graphical tool for acquiring temporal-abstraction knowledge is developed using the PROTEGE-II system. An evaluation is being performed of the knowledge-acquisition tool, the computational mechanisms, and the overall framework in the diabetes-therapy domain and within the EON architecture for guideline-based care. The research has implications for construction of knowledge-based planning and monitoring systems in time-oriented domains, for automated summarization of temporal databases, for the operational semantics of deductive temporal databases, and for the semantics of queries to temporal databases; preliminary evaluation in clinical and engineering domains is highly encouraging.
